Examining Post Acute Care Relationships in an Integrated Hospital System. 2.4.3 Multivariate Analyses

02/01/2009

Five multivariate models were run to examine the effects of beneficiary demographics, supply of PAC providers, severity of illness, and organizational relationships on any post-acute service use, index admission length of stay, first site of PAC, acute hospital readmission during post-acute episodes, and total Medicare payments for episodes of PAC.

The models predicting post-acute service use and readmission during post-acute episodes were binomial logistic regression models in which the dependent variables were the presence or absence of a post-acute episode claim (1/0) or an acute hospital readmission during the episode of care. The models predicting index admission length of stay and total Medicare payments for episodes of post-acute care were ordinary least squares (OLS) regression models with a continuous dependent variable indicating the beneficiary's length of stay in the index hospitalization and the Medicare payment amount for the PAC episode. The fifth model was a multinomial logistic regression model predicting the first site of post-acute care for the subset of hospital discharges with a post-acute episode claim. The reference group for this model was hospital outpatient therapy meaning that all coefficients generated from the model are interpreted in comparison to beneficiaries discharged to hospital outpatient therapy. The model predicts the odds of being discharged to SNF, HHA, IRF, or LTCH compared to being discharged to hospital outpatient therapy.

These five models were run using three sets of independent variables. These sets of variables differ based on the severity measures. Table 2-1 contains the variables used in the multivariate models and highlights the differences in the use of severity measures across modules.

The independent variables across models included demographic characteristics such as gender, Medicaid status, age, and race; severity of illness level; supply of IRF, SNF, and LTCH beds per state; and census division. The supply measures of beds per beneficiary per state were included to control for availability of PAC providers and potential provider substitution. These measures are based on 2007 Medicare POS. Home health agencies and hospital outpatient departments are widely available across the nation and were not identified in these models. Characteristics of the discharging acute hospital were also included in the models predicting any post-acute service use, acute admission length of stay, readmission during a post-acute episode, and post-acute care episode payments.

These variables included number of beds in the acute hospital, urban versus rural location, and for-profit versus not-for-profit versus government-run control.

Table 2-1. Independent Variables for Multivariate Analysis
Variable Names Independent Variables
Set 1 Set 2 Set 3
Demographics Female X X X
Any Medicaid in 2005 X X X
Aged 65-74 X X X
Aged 75-84 X X X
Aged 85+ X X X
Post-Acute Care Supply Variables IRF beds/1,000 beneficiaries/state X X X
SNF beds/1,000 beneficiaries /state X X X
LTCH beds/1,000 beneficiaries /state X X X
Census Division Indicators Middle Atlantic X X X
East North Central X X X
West North Central X X X
South Atlantic X X X
East South Central X X X
West South Central X X X
Mountain X X X
Pacific X X X
Organizational Relationships Variables Presence of colocated providers X X X
Presence of subproviders X X X
Characteristics of Discharging Acute Hospital Number of Beds X X X
Urban Location X X X
Not-for-Profit X X X
Government Owned X X X
Index Acute DRG DRG 209 544 Major Joint & Limb Reattachment Procedures of Lower Extremity X X  
DRG 089 Simple Pneumonia & Pleurisy Age > 17 w CC X X  
DRG 014 Specific Cerebrovascular Disorders Except TIA X X  
DRG 127 Heart Failure & Shock X X  
DRG 210 Hip & Femur Procedures Except Major Joint Age > 17 w CC X X  
APR-DRG Severity Measure APR-DRG Severity Index = moderate X    
APR-DRG Severity Index = major X    
APR-DRG Severity Index = extreme X    
MS-DRG Severity Measure MS-DRG Severity Index = CC   X X
MS-DRG Severity Index = MCC   X X
Hierarchical Condition Category Indicators HCC80 Congestive Heart Failure     X
HCC92 Specified Heart Arrhythmias     X
HCC108 Chronic Obstructive Pulmonary Disease     X
HCC19 Diabetes without Complication     X
HCC131 Renal Failure     X
HCC79 Cardio-Respiratory Failure and Shock     X
HCC158 Hip Fracture/Dislocation     X
HCC164 Major Complications of Medical Care and Trauma     X
HCC105 Vascular Disease     X
HCC96 Ischemic or Unspecified Stroke     X
Demographics Female X X X
Any Medicaid in 2005 X X X
Aged 65-74 X X X
Aged 75-84 X X X
Aged 85+ X X X
Post-Acute Care Supply Variables IRF beds/1,000 beneficiaries/state X X X
SNF beds/1,000 beneficiaries /state X X X
LTCH beds/1,000 beneficiaries /state X X X
HCC83 Angina Pectoris/Od Myocardial Infarction     X

Indicator variables for the five most frequent hospital DRGs for PAC users were included in the first two sets of independent variables to capture the added effect of particular diagnoses and the impact of medical versus rehabilitation DRGs in each of the models. These top five DRGs included DRG 544: Major Joint & Limb Reattachment Procedures of Lower Extremity; DRG 089: Simple Pneumonia & Pleurisy Age > 17 w CC; DRG 014: Specific Cerebrovascular Disorders Except TIA; DRG 127: Heart Failure & Shock; and DRG 210: Hip & Femur Procedures Except Major Joint Age > 17 w CC. These five DRGs encompass the most common three DRGs in each PAC setting except for LTCHs. The two most common LTCH DRGs are for tracheostomy procedures and these DRGs are discharged to non-LTCH settings with very low frequency. The uncommon observance of these DRGs in the other PAC settings led to model convergence issues when these DRGs were included.

Organizational relationship variables were also included in each multivariate model. The models predicting any PAC use, acute hospital readmission, PAC episode payment, and predicting acute index admission length of stay included dummy variables indicating if the acute index hospital had any type of subprovider or any type of colocated provider. The multinomial logit model included more specific organizational relationship variables indicating the presence or absence of specific post-acute subproviders or colocated providers. Because the multinomial logit predicts the specific setting of PAC, it is important to know whether the acute index hospital has an organizational relationship with the type of post-acute provider to which a beneficiary is discharged. Correlations were run for all independent variables included in the multivariate models and no significant correlation was noted.

As indicated in Table 2-1, the three sets of independent variables used in the multivariate analyses varied in terms of the severity measures included. The APR-DRG severity measures were used in order to compare the results of the regressions using the 2006 Medicare claims data to the results using 2005 Medicare claims data as reported in last year's findings. In the second set of independent variables, we replaced the APR-DRG severity variables with the MS-DRG severity variables in order to reflect current CMS policy which has mandated the use of MS-DRGs as of FY 2008. Given that the MS-DRGs have fewer categories that measure differences in clinical severity of illness, we also ran a third set of independent variables to include HCC indicator variables along with the MS-DRGs. In these models we removed the indicators for the five most frequent hospital DRGs due to high correlation of these variables to the HCCs. We included HCC indicators variables for HCCs present in 5.0 percent of beneficiaries in the sample. These HCCs included:

  • HCC80 Congestive Heart Failure
  • HCC92 Specified Heart Arrhythmias
  • HCC108 Chronic Obstructive Pulmonary Disease
  • HCC19 Diabetes without Complication
  • HCC131 Renal Failure
  • HCC79 Cardio-Respiratory Failure and Shock
  • HCC158 Hip Fracture/Dislocation
  • HCC164 Major Complications of Medical Care and Trauma
  • HCC105 Vascular Disease
  • HCC96 Ischemic or Unspecified Stroke
  • HCC83 Angina Pectoris/Old Myocardial Infarction

View full report

Preview
Download

"report.pdf" (pdf, 1.14Mb)

Note: Documents in PDF format require the Adobe Acrobat Reader®. If you experience problems with PDF documents, please download the latest version of the Reader®